2019 Volume 69 Issue 3 Pages 420-428
Yield improvement is a top priority for maize breeding. Kernel size and weight are important determinants of maize grain yield. In this study, a recombinant inbred line (RIL) population and an association panel were used to identify quantitative trait loci (QTLs) for four maize kernel-related traits: kernel length, width, thickness and 100-kernel weight. Twenty-seven QTLs were identified for kernel-related traits across three environments and the best linear unbiased predictions (BLUPs) of each trait by linkage analysis, and four QTLs were stably detected in more than two environments. Additionally, 29 single nucleotide polymorphisms (SNPs) were identified as significantly associated with the four kernel-related traits and BLUPs by genome-wide association study, and two loci could be stably detected in both environments. In total, four QTLs/SNPs were co-associated with various traits in both populations. Using combined-linkage analysis and association mapping, PZE-101066560 on chromosome 1, associated with kernel width and with 100-kernel weight in the association panel, was co-localized within the QTL interval of qKW1-3 for kernel width in the RILs. Two annotated genes in the candidate region were considered as potential candidate genes. The QTLs and candidate genes identified here will facilitate molecular breeding for grain yield improvement in maize.
Maize (Zea mays L.) is an important crop for food, feed, and fuel worldwide, and improving grain yield is a top priority in current maize breeding programs (Li et al. 2011b, Qin et al. 2016, Su et al. 2017). Maize grain yield is determined by several secondary components, among which the kernel size, as assessed by kernel length (KL), kernel width (KW), and kernel thickness (KT), plays a key role in determining kernel weight, and thus grain yield (Chen et al. 2016c, Gupta et al. 2006, Liu et al. 2014, Raihan et al. 2016). The kernel size also affects grain filling, as well as seedling vigor and end-use quality, which can influence the market grade and consumer preference (Chen et al. 2016a, Liu et al. 2011, Raihan et al. 2016, Revilla et al. 1999). The maize kernel-related traits are classic quantitative traits with complex genetic mechanisms that are under the control of quantitative trait loci (QTLs) with small effects and under the influence of environmental changes (Chen et al. 2016b, Qin et al. 2016, Raihan et al. 2016). Thus, understanding the genetic mechanisms of maize kernel-related traits is critical for the genetic manipulation of grain yield.
With the advent of crop genomics and development of molecular markers, QTL-based approaches have been proven to be powerful tools for elucidating the genetic basis of kernel-related traits, thus allowing improvements in high-yield breeding efficiency (Chen et al. 2016b, Li et al. 2011a, Monaco et al. 2013, Raihan et al. 2016). In rice, several genes (or QTLs) for kernel-related traits have been isolated and functionally characterized using map-based cloning (Huang et al. 2013, Li et al. 2011a), including GS2 (Zhang et al. 2013), GS3 (Fan et al. 2006), GS5 (Li et al. 2011a), GL3.1 (Qi et al. 2012), GW2 (Song et al. 2007), GW5 (Weng et al. 2008), GW7 (Wang et al. 2015), and GW8 (Wang et al. 2012). In maize, genes directly affecting kernel-related traits were rarely identified through natural genetic variation (Chen et al. 2016c), but several genes for kernel-related traits, such as gln1-3/gln1-4 (Martin et al. 2006), rgf1 (Maitz et al. 2000), dek1 (Lid et al. 2002), sh1 and sh2 (Thévenot et al. 2005), and incw2 (Carlson and Chourey 1999), were isolated using maize mutants. In addition, four orthologues of rice GS3, GS5 and GW2, namely ZmGS3, ZmGS5, and ZmGW2-CHR4 and ZmGW2-CHR5, respectively, were isolated from maize using homology-based cloning, and determined to be associated with kernel-related traits (Li et al. 2010a, 2010b, Liu et al. 2015). These genes facilitated the dissection of kernel development and its regulation (Chen et al. 2016c). Compared with mutant effects and homology-based cloning, the QTL-based mapping approach is ideal to detect favorable QTLs/genes underlying the natural variations in kernel-related traits (Chen et al. 2016a). Considerable effort has been applied to dissect the genetic variation of grain yield, and many QTLs associated with kernel-related traits in maize have been identified (Chen et al. 2016a, 2016b, 2016c, Li et al. 2013, 2016, Liu et al. 2014, Ma et al. 2007, Messmer et al. 2009, Qin et al. 2016, Raihan et al. 2016, Su et al. 2017, Yang et al. 2016, Zhang et al. 2014, 2017, Zhou et al. 2017). However, such QTLs, mainly identified by linkage analysis based on bi-parental populations, were only partially consistent across various studies using different populations, suggesting the strong influence of the genetic background. In addition, because of the large confidence intervals of the QTLs (caused by the limited recombination events) and the restricted alleles in the bi-parental genotypes, very few QTLs have been incorporated into maize high-yield breeding programs (Hao et al. 2015, Riedelsheimer et al. 2012).
The technical advances in next-generation sequencing and the development of improved statistical methods have enabled using genome-wide association studies (GWAS) based on linkage disequilibrium (LD) analyses as powerful alternatives for examining quantitative traits in plants (Xiao et al. 2017). Compared with linkage mapping based on bi-parental populations, GWAS has the potential to exploit nearly all recombination events in the evolutionary history of a specific germplasm, providing increased map resolution, and it can simultaneously evaluate the varying effects of multiple alleles (Lu et al. 2010). The key constraints for the successful use of GWAS in plants are related to population structure and rare alleles (low-frequency functional alleles), which severely limit the power of QTL detection (Gupta et al. 2005, Lu et al. 2010). However, the combination of linkage mapping and GWAS is able to effectively overcome some of the inherent limitations of each method and has been successfully applied to identify QTLs underlying drought tolerance, male inflorescence size, and grain morphology in maize (Li et al. 2016, Lu et al. 2010, Zhang et al. 2017).
In the present study, a maize recombinant inbred line (RIL) population, derived from ‘DH1M’ × ‘T877’, and a set of 253 elite maize inbred lines from modern breeding programs were evaluated to detect the QTLs underlying maize kernel-related traits based on high-density skeleton bin map, and single nucleotide polymorphism (SNP) loci associated with kernel-related traits using GWAS. Consistent QTLs for kernel-related traits were identified by combining linkage mapping and GWAS. Information on the QTLs or SNP loci involved in the kernel-related traits identified in this study will facilitate breeding strategies for grain yield improvement in maize.
The RIL population used here consisting of 204 lines (Li et al. 2018) produced by single-seed decent, and derived from a cross between two elite maize inbred lines, ‘DH1M’ and ‘T877’, in Chinese maize breeding programs (Xue et al. 2010, Zhao et al. 2001). These two parental lines differ significantly in kernel size and grain yield, thus providing materials to examine the genetic basis of maize kernel-related traits. The 204 RILs and the two parental lines were evaluated in Nantong and Sanya (both in China) in 2016 (Nantong only) and 2017. Each location plus year combination was considered as an environment: Nantong 2016 and 2017 were designated as environments E1 and E2, respectively, while Sanya 2017 was designated as E3.
A set of 253 elite maize inbred lines with abundant phenotypic variation, including traditional landraces and improved maize inbred lines selected from a wide range of geographical locations in China (Supplemental Table 1), were collected to construct an association panel for GWAS (Zhou et al. 2016a). This association panel was evaluated in Nantong in 2016 and 2017, which were designated as environment E1′ (Nantong 2016) and E2′ (Nantong 2017), respectively.
Two populations were arranged following a randomized complete block design with three replicates in all environments. Each genotype was grown in a single row (500-cm long; 60-cm between rows), each with 20 plants, and standard agronomic practices were performed in each environment. Five representative well opened pollinated ears per line and per replicate were harvested for phenotypic measurements after maturity. Four kernel-related traits, namely kernel length (KL, mm), kernel width (KW, mm), kernel thickness (KT, mm), and 100-kernel weight (HKW, g), were evaluated in all environments. KL, KW, and KT were estimated from the averages of 10 randomly selected kernels from the center of each ear using an electronic digital caliper. Whereas HKW was measured as the average weight of three samples of 100 mixed seeds from five ears.
Phenotypic data analysesStatistical analyses of all phenotypic data of the tested traits across different environments were performed using SAS 9.13 software (SAS Institute, Cary, NC, USA). The broad-sense heritability (H2) of each trait was estimated according to the formula: H2 (%) = σ2g/(σ2g + σ2ge/n + σ2e/nr) × 100%, where σ2g is the genotypic variance, σ2ge is the variance for interactions of genotype with environment, σ2e is the error variance, n is the number of environments, and r is the number of replications (Hallauer et al. 2010), which were estimated using the PROC VARCOMP procedure in SAS. To minimize the effects of environmental factors, the best linear unbiased predictions (BLUPs) for each trait in each maize inbred line across environments were evaluated using the PROC MIXED procedure in SAS. Pearson correlation coefficients between the tested traits were calculated using the PROC CORR procedure in SAS, based on the BLUPs of the traits across environments.
Linkage map construction and QTL mappingThe ‘DH1M’/‘T877’ RILs and their parents were genotyped using an Affymetrix CGMB50K SNP Array containing 56,000 maize SNPs at China Golden Marker (Beijing) Biotech, China, and polymorphisms were explored using 9,780 SNP markers. A high-density genetic map was constructed for all 204 RILs using 1,868 genetic bins, covering 3,081.8 cM of mapping distance, with an average genetic distance of 1.65 cM between adjacent bin markers (Li et al. 2018). Based on this high-density genetic map, a QTL analysis was performed in IciMapping v4.1 (Lei et al. 2015) using the inclusive composite interval mapping (ICIM) method. The scanning step size was set to 1 cM, and the largest P value for entering variables in the stepwise regression of phenotype on marker variables (PIN) was set at 0.001. For all traits examined, the logarithm of odds (LOD) threshold was determined using 1000 permutations at a significance level of P = 0.05. QTLs with LOD values larger than the threshold value (threshold = 3.36 after 1000 permutations) were considered further.
GWASThe association panel of 253 maize inbred lines was previously genotyped (Zhou et al. 2016a) using a Maize SNP3K Beadchip (Illumina, San Diego, CA, USA) at the National Maize Improvement Centre of China, China Agricultural University, Beijing, China. The SNP3K Beadchip contains 3,072 random SNPs, evenly covering the maize genome, that were selected from a maize SNP genotyping array (Ganal et al. 2011, Hao et al. 2015). The 2,824 SNPs with missing data ≤ 20% and minor allele frequency (MAF) ≥ 5% were used for further analyses.
The population structure was evaluated in STRUCTURE 2.3 software (Pritchard et al. 2000) using the 2,824 SNPs. The 253 maize inbred lines were previously divided into two major subpopulations (Zhou et al. 2016a), and the corresponding Q matrix was used for the GWAS performed in this study. To estimate the genetic relatedness among individuals, the relative kinship matrix (K) was calculated in SPAGeDi software (Hardy and Vekemans 2002) using the 2,824 SNPs (Zhou et al. 2016a). The GWAS was performed in TASSEL4.0 software (Bradbury et al. 2007), and a mixed linear model (MLM, Q + K) was applied to account for the population structure (Q) and relative kinship (K) using previously defined Q and K variables (Deng et al. 2017, Yu et al. 2006, Zhou et al. 2016a). Initially, Bonferroni correction and simpleM method (Gao et al. 2010) were used to correct multiple hypothesis testing. However, no SNP was identified in both environments and across environments, indicating many true positive loci might be lost. So, a reliable measure of significance of positive false discovery rate (FDR) method was additionally applied to correct for multiple testing using QVALUE software in R (Storey and Tibshirani 2003).
Phenotype variations of kernel-related traits in the two mapping populations are presented in Tables 1 and 2. All four tested traits varied widely in both the RIL and the association mapping populations. For example, the BLUPs for KW ranged from 6.69 to 9.78 mm in the RIL population, and from 6.42 to 10.76 mm in the association mapping population; the BLUPs for HKW ranged from 14.88 to 31.87 g in the RIL population, and from 8.68 to 31.60 g in the association mapping population. An analysis of variance for all tested kernel-related traits suggested that the genotypic and genotype-by-environmental effects were highly significant at the 0.001 probability level in both populations. The estimated H2 of the kernel-related traits ranged from 90.52 for KT to 96.47 for KW in the RIL population, and from 79.34 for KL to 93.44 for KW in the association mapping population.
| Traitsa | Env.b | RIL population | Parental lines (mean ± SD) | Gc | G × Ed | H2e (%) | ||
|---|---|---|---|---|---|---|---|---|
| Mean ± SD | Range | T877 | DH1M | |||||
| KL (mm) | E1 | 9.08 ± 0.68 | 7.08–11.45 | 8.63 ± 0.09 | 9.40 ± 0.34 | ** | ** | 92.74 |
| E2 | 9.09 ± 0.56 | 7.12–11.01 | 9.29 ± 0.22 | 9.78 ± 0.21 | ||||
| E3 | 9.33 ± 0.68 | 7.09–11.60 | 8.88 ± 0.17 | 9.55 ± 0.24 | ||||
| BLUPs | 9.16 ± 0.55 | 7.23–11.19 | 8.94 ± 0.33 | 9.58 ± 0.19 | ||||
| KW (mm) | E1 | 8.18 ± 0.57 | 6.54–10.04 | 8.34 ± 0.08 | 8.91 ± 0.15 | ** | ** | 96.47 |
| E2 | 8.11 ± 0.50 | 6.63–9.69 | 8.41 ± 0.10 | 8.80 ± 0.15 | ||||
| E3 | 8.32 ± 0.55 | 6.62–9.81 | 8.43 ± 0.15 | 8.96 ± 0.07 | ||||
| BLUPs | 8.21 ± 0.49 | 6.69–9.78 | 8.40 ± 0.05 | 8.89 ± 0.08 | ||||
| KT (mm) | E1 | 5.21 ± 0.56 | 3.76–6.72 | 5.12 ± 0.13 | 4.48 ± 0.20 | ** | ** | 90.52 |
| E2 | 5.09 ± 0.48 | 4.00–6.85 | 4.99 ± 0.19 | 4.56 ± 0.27 | ||||
| E3 | 5.09 ± 0.51 | 3.76–6.69 | 4.94 ± 0.25 | 4.48 ± 0.20 | ||||
| BLUPs | 5.13 ± 0.45 | 3.99–6.64 | 5.02 ± 0.09 | 4.50 ± 0.05 | ||||
| HKW (g) | E1 | 22.61 ± 3.58 | 13.27–34.50 | 18.23 ± 0.65 | 21.12 ± 0.89 | ** | ** | 95.04 |
| E2 | 22.60 ± 3.30 | 13.73–32.47 | 19.91 ± 0.18 | 22.82 ± 0.90 | ||||
| E3 | 23.52 ± 3.53 | 13.27–32.67 | 18.20 ± 0.89 | 21.37 ± 0.42 | ||||
| BLUPs | 22.92 ± 3.11 | 14.88–31.87 | 18.78 ± 0.98 | 21.77 ± 0.92 | ||||
| Traitsa | Env.b | Mean ± SD | Range | Gc | G × Ed | H2 (%)e |
|---|---|---|---|---|---|---|
| KL(mm) | E1′ | 8.82 ± 1.01 | 6.46–12.96 | ** | ** | 79.34 |
| E2′ | 9.16 ± 0.81 | 6.22–11.62 | ||||
| BLUPs | 9.00 ± 0.73 | 7.14–11.85 | ||||
| KW(mm) | E1′ | 8.08 ± 0.81 | 5.98–11.96 | ** | ** | 93.44 |
| E2′ | 8.25 ± 0.70 | 5.86–10.50 | ||||
| BLUPs | 8.18 ± 0.64 | 6.42–10.76 | ||||
| KT(mm) | E1′ | 4.64 ± 0.54 | 3.13–7.23 | ** | ** | 84.10 |
| E2′ | 4.76 ± 0.49 | 3.33–6.65 | ||||
| BLUPs | 4.70 ± 0.38 | 3.84–6.09 | ||||
| HKW(g) | E1′ | 19.09 ± 5.91 | 5.48–34.15 | ** | ** | 92.72 |
| E2′ | 21.09 ± 4.08 | 8.42–31.54 | ||||
| BLUPs | 20.14 ± 4.28 | 8.68–31.60 |
The four traits in the two populations were significantly correlated with each other, except for KL and KT in the RIL population (Table 3). Significant positive correlations were observed between HKW and the other three traits, with correlation coefficient (r) values of 0.400, 0.631, and 0.433 for KL, KW, and KT, respectively, in the RIL population, and of 0.516, 0.744, and 0.589 for KL, KW, and KT, respectively, in the association mapping population.
| Traitsa | KL | KW | KT | HKW |
|---|---|---|---|---|
| KL | 0.448** | −0.057 | 0.400** | |
| KW | 0.440** | 0.201** | 0.631** | |
| KT | 0.194* | 0.515** | 0.433** | |
| HKW | 0.516** | 0.744** | 0.589** |
Correlation coefficients for the RIL population are above the diagonal, while those for the association panel are below the diagonal.
Twenty-seven QTLs were detected for the four kernel-related traits across the three different environments and the BLUPs across all of the environments, which were distributed on eight maize chromosomes, excluding chromosome 7 and 9 (Table 4). In total, three QTLs underlying KL were located on chromosomes 3, 5, and 8. The 15 QTLs for KW were distributed on seven chromosomes, but none were found on chromosomes 7, 8, and 9. The 9 QTLs for HKW were located on chromosome 1, 2, 3, 4, and 5. However, no QTLs were detected for KT. The percentage of phenotypic variance explained by the individual QTLs ranged from 3.34% (KW, qKW2-2) to 11.84% (KW, qKW10-1) for the different traits (Table 4). Of these QTLs, 14 (~50.0%) showed positive additive effects, which indicated that the favorable alleles from ‘DH1M’ could increase the values of the traits in the corresponding environments.
| Traitsa | QTLb | Chr | Position (cM) | Marker interval | E1 | E2 | E3 | BLUPs | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LODc | PVEd (%) | Adde | LOD | PVE (%) | Add | LOD | PVE (%) | Add | LOD | PVE (%) | Add | |||||
| KL | qKL3-1 | 3 | 125 | PZE-103049569_PZE-103051543 | 3.41 | 5.43 | −0.13 | |||||||||
| qKL5-1 | 5 | 183 | SYN35495_PZE-105110743 | 6.06 | 10.23 | −0.18 | ||||||||||
| qKL8-1 | 8 | 157 | PZE-108069615_PZE-108070036 | 3.74 | 5.96 | 0.14 | ||||||||||
| KW | qKW1-1 | 1 | 27 | SYN7706_SYN6742 | 6.35 | 6.59 | −0.16 | 5.77 | 6.20 | −0.13 | ||||||
| qKW1-2 | 1 | 61 | SYN33163_SYN367 | 7.55 | 8.30 | 0.18 | 8.86 | 10.33 | 0.17 | |||||||
| qKW1-3 | 1 | 121 | PZE-101066217_SYN29479 | 4.82 | 4.86 | 0.14 | 5.54 | 8.08 | 0.14 | 5.65 | 8.38 | 0.16 | 4.45 | 4.61 | 0.11 | |
| qKW2-1 | 2 | 87 | SYN451_PZE-102039760 | 5.81 | 6.34 | 0.13 | ||||||||||
| qKW2-2 | 2 | 118 | SYNGENTA3962_PZE-102065594 | 3.40 | 3.34 | 0.11 | ||||||||||
| qKW2-3 | 2 | 145 | PZE-102108955_PZE-102109699 | 6.91 | 10.31 | 0.18 | ||||||||||
| qKW2-4 | 2 | 167 | PZE-102126983_SYN33456 | 3.58 | 5.12 | 0.11 | ||||||||||
| qKW3-1 | 3 | 5 | PZE-103000497_PZE-103001626 | 3.60 | 4.18 | 0.13 | 4.44 | 5.35 | 0.12 | |||||||
| qKW4-1 | 4 | 43 | PZE-104010039_PZE-104010477 | 3.40 | 3.47 | 0.12 | 5.12 | 7.79 | 0.16 | 3.75 | 3.98 | 0.11 | ||||
| qKW4-2 | 4 | 194 | PZE-104087825_PZE-104088242 | 4.91 | 7.17 | −0.15 | ||||||||||
| qKW5-1 | 5 | 129 | PZE-105062861_SYN1318 | 5.96 | 6.12 | −0.15 | ||||||||||
| qKW5-2 | 5 | 184 | PZE-105111506_PZE-105113106 | 3.85 | 3.96 | −0.10 | ||||||||||
| qKW6-1 | 6 | 39 | PZE-106014381_PZE-106014738 | 3.38 | 4.86 | −0.11 | 3.98 | 4.15 | −0.11 | |||||||
| qKW10-1 | 10 | 98 | SYN22128_PZE-110051617 | 10.89 | 11.84 | −0.22 | 9.14 | 10.04 | −0.17 | |||||||
| qKW10-2 | 10 | 101 | PZE-110054072_PZE-110054977 | 6.38 | 9.38 | −0.15 | ||||||||||
| HKW | qHKW1-1 | 1 | 60 | SYN33163_SYN367 | 5.67 | 8.97 | 1.14 | 5.13 | 6.99 | 0.88 | 5.38 | 9.46 | 1.10 | 4.60 | 7.39 | 0.87 |
| qHKW1-2 | 1 | 296 | SYN15632_PZE-101178511 | 3.90 | 5.21 | 0.77 | ||||||||||
| qHKW2-1 | 2 | 98 | SYN12624_SYN635 | 5.68 | 8.77 | 1.13 | 4.33 | 7.56 | 0.99 | 6.22 | 10.08 | 1.03 | ||||
| qHKW2-2 | 2 | 124 | SYN29649_SYN23572 | 5.18 | 7.03 | 0.89 | ||||||||||
| qHKW3-1 | 3 | 1 | PZE-103000190_PZE-103000497 | 3.77 | 5.03 | 0.75 | ||||||||||
| qHKW4-1 | 4 | 243 | PZE-104118348_PZE-104119147 | 3.46 | 5.60 | −0.87 | ||||||||||
| qHKW4-2 | 4 | 269 | SYN30108_PZE-104139562 | 4.28 | 6.21 | −0.84 | ||||||||||
| qHKW5-1 | 5 | 181 | PZE-105107790_PZE-105108927 | 4.57 | 6.39 | −0.84 | 3.71 | 5.72 | −0.77 | |||||||
| qHKW5-2 | 5 | 210 | PZE-105130823_PZE-105132493 | 4.00 | 5.73 | −0.91 | ||||||||||
Among these detected QTLs, only four (designated as ‘consistent’ QTLs) were stably detected in two or more environments as well as across environments (Table 4). Of these consistent QTLs, two were stably identified for KW (on chromosomes 1 and 4) and two were stably identified for HKW (on chromosomes 1 and 2). However, no consistent QTLs were detected for KL or KT. A QTL for KW (qKW1-3 located in the marker interval of PZE-101066217_SYN29479 on chromosome 1) and a QTL for HKW (qHKW1-1 flanked by SYN33163 and SYN367 on chromosome 1) were stably identified in all different environments and across environments. The positive additive effects of these consistent QTLs for KW and HKW were contributed by the favorable alleles from the ‘DH1M’ parent, which had large KW and HKW.
Among these 27 QTLs, two QTLs of qHKW1-1 for HKW and qKW1-2 for KW were co-localized in the same marker interval of SYN33163 and SYN367 on chromosome 1.
GWAS of kernel-related traitsOverall, 2,824 high-quality SNPs with missing data ≤ 20% and MAF ≥ 5% in the association mapping population were used for the GWAS performed here. Based on the MLM incorporating Q + K, 461 marker-trait associations involving 231 SNPs, representing all chromosomes, were identified as being associated with the four kernel-related traits in different environments and the BLUPs across environments at the threshold level of P ≤ 0.01 (–LogP ≥ 2.00) (Supplemental Figs. 1, 2). After correction for multiple testing using the FDR method (FDR ≤ 0.05), 40 marker-trait associations representing 29 SNPs remained significant, these SNPs were distributed among 8 chromosomes, excluding chromosome 5 and 9 (Supplemental Table 2). Of these 29 SNPs, nine were significantly associated with KW, KT, and HKW, in at least one environment as well as the BLUPs across environments (Table 5). Two SNPs of PZE-107099124 on chromosome 7 and PZE-110017983 on chromosome 10 were stably identified for KW and HKW, respectively, in both environments and across environments.
| Traita | Marker | Chr.b | Positionc (b) | P value | q valued | ||||
|---|---|---|---|---|---|---|---|---|---|
| E1′ | E2′ | BLUPs | E1′ | E2′ | BLUPs | ||||
| KW | PZE-103033919 | 3 | 26447512 | 0.00020 | 0.00058 | 0.03174 | 0.03624 | ||
| PZE-106041751 | 6 | 91150330 | 0.00110 | 0.00063 | 0.03176 | 0.04119 | |||
| PZE-107099124 | 7 | 154179180 | 0.00015 | 0.00080 | 0.00010 | 0.03026 | 0.01313 | 0.02381 | |
| KT | PZE-106049961 | 6 | 99771219 | 0.00027 | 0.00036 | 0.01918 | 0.02849 | ||
| HKW | PZE-101066560 | 1 | 49945398 | 0.00104 | 0.00127 | 0.04715 | 0.04824 | ||
| PZE-103049396 | 3 | 53675536 | 0.00090 | 0.00136 | 0.04102 | 0.03609 | |||
| PZE-106012837 | 6 | 32842201 | 0.00078 | 0.00065 | 0.02238 | 0.03684 | |||
| PZE-108103023 | 8 | 158762013 | 0.00025 | 0.00028 | 0.02196 | 0.02245 | |||
| PZE-110017983 | 10 | 20712573 | 0.00055 | 0.00031 | 0.00032 | 0.02519 | 0.03390 | 0.03493 | |
After correction for multiple testing using the FDR method (FDR ≤ 0.05), two SNPs showed pleiotropy with different traits in this association panel (Supplemental Table 2), PZE-101066560 on chromosome 1 and PZE-103033919 on chromosome 3 were co-associated with KW and HKW.
Co-localization of QTLs for kernel-related traits by combined linkage mapping and GWASThe QTLs for kernel-related traits identified in the RILs were compared with the GWAS results in the association panel. The SNP PZE-101066560, located within a 165-kb LD block on chromosome 1 (Zhou et al. 2016a), which was identified as being associated with KW and HKW by GWAS, was found within the qKW1-3 interval for KW in the RILs (Tables 4, 5, Supplemental Table 2), while other QTLs for kernel-related traits were not co-localized according to the combined linkage mapping and GWAS. Candidate genes in the genomic interval of qKW1-3 (flanked by PZE-101066217 and SYN29479) on chromosome 1 were predicted based on the B73 reference genome sequence (http://www.maizegdb.org). The genomic interval of PZE-101066217_SYN29479 was approximately 1.4 Mb, and 23 protein-encoding genes were located in this genomic region (Supplemental Table 3). According to the maize gene annotation database at MaizeGDB (http://www.maizegdb.org), five candidate genes (GRMZM2G092442, GRMZM2G043484, GRMZM5G824629, GRMZM2G056373 and GRMZM2G011483) were located in the LD block of PZE-101066560 (within 330 kb, 165-kb upstream and downstream of the SNP PZE-101066560), and two annotated genes (GRMZM5G824629 and GRMZM2G011483) were the most likely candidate genes for qKW1-3. The candidate gene GRMZM5G824629 encodes a ubiquitin-conjugating enzyme, and the candidate gene GRMZM2G011483 encodes a cyclin-related protein (Supplemental Table 3).
Yield improvement is a top priority for maize breeding (Chen et al. 2016b). Kernel size and kernel weight are important components of maize yield, and the latter is largely determined by kernel size, which has been strongly selected during maize domestication and improvement (Gupta et al. 2006, Liu et al. 2017, Raihan et al. 2016). In the present study, wide variations in kernel-related traits were observed in both the RIL population and the association panel, and significant positive correlations were also observed between HKW and the other three kernel size traits in both populations. This indicated that kernel size plays an important role in determining HKW, suggesting that these traits might be simultaneously improved in breeding programs (Zhang et al. 2017). Moreover, the correlation coefficients for KW were usually greater than those for KL and KT in both evaluated populations, as found by Liu et al. (2017), suggesting that KW might play a more important role than other traits in determining kernel weight in maize. The estimates of H2 were high for all kernel-related traits in both evaluated populations, suggesting that genetic factors play an important role in the formation of these traits, which was consistent with the results of previous studies (Liu et al. 2014, Raihan et al. 2016). In addition, obvious bi-directional transgressive segregation was detected for all traits in the RIL population, indicating their polygenic control. It is important to map QTLs for kernel-related traits to increase our understanding of the genetic and molecular bases of maize grain yield, which will facilitate marker-assisted selection (MAS) to select for the genetic determinant(s) of maize yield.
In the present study, linkage mapping and GWAS were used to dissect the candidate loci associated with the kernel-related traits of maize. Twenty-seven QTLs and 29 associated SNPs were identified for the four kernel-related traits examined here through linkage mapping and GWAS, respectively. Most of the identified QTLs or loci with small effects confirmed that the kernel-related traits in modern maize inbred lines are controlled by multiple genes with low effects, in agreement with the results of a previous study (Zhang et al. 2017). Only four of these QTLs for kernel-related traits were stably detected in at least two environments or across environments in the RIL population, and only two loci could be stably detected in both environments and across environments in the association panel, indicating that many of the QTLs/SNPs detected were environmentally specific. Thus, most QTLs for kernel-related traits of maize might be affected by environmental factors, having QTL-by-environment interactions, and controlled by multiple genes with minor effects (Hao et al. 2017, Raihan et al. 2016, Zhou et al. 2016b). However, only stable and highly heritable QTLs will be useful for the MAS aiming to increase maize yield in a wide range of environments in breeding programs (Zhou et al. 2016b). The stable QTLs identified in the present study for four kernel-related traits that have relatively high heritability levels, such as qKW1-3 and qHKW1-1, should be considered priority candidates for MAS in modern maize breeding programs. Some of the four stable QTLs identified here were located in the same regions as those identified for yield-related traits in previous studies, such as qKW1-3 for KW in bin 1.02 (Liu et al. 2014), qKW4-1 for KW in bin 4.02 (Austin and Lee 1996), and qHKW1-1 for HKW in bin 1.02 (Liu et al. 2014). The stable QTL of qHKW2-1 is a novel QTL. These results indicated that some causal gene/genes might be located in these regions, and these stable QTLs might be further used for fine mapping, gene cloning, and validation of the potential candidate genes, which may be highly valuable in maize breeding (Zhang et al. 2017, Zhou et al. 2016b).
The QTLs for domestication-related traits form clusters that are consistent with the regions harboring favorable genes (Cai and Morishima 2002, Liu et al. 2014, Zhang et al. 2017). In the present study, two QTLs from RILs and two SNPs from the association panel were associated with various traits, which shared significant correlations. Some of these QTL/SNPs were located in or near regions where QTLs for yield-related traits have been mapped in previous studies (Agrama and Moussa 1996, Liu et al. 2014). For example, a QTL cluster in the interval of SYN33163_SYN367 on chromosome 1 was responsible for two significant QTLs, qKW1-2 and qHKW1-1, in the RILs, the SNP PZE-101066560 on chromosome 1 was co-associated with KW and HKW in the association panel. The clustering or co-association of QTLs/SNPs for different kernel-related traits in the two evaluated populations could be explained by the pleiotropy of the same gene(s) or by the joint effects of closely linked genes (in local LD) in the identified regions associated with target traits, which may benefit from the association of adaptive phenotypes during domestication (Bergelson and Roux 2010, Liu et al. 2014, Marathi et al. 2012, Zhang et al. 2017). In modern breeding schemes for maize grain yield improvement, MAS of these clustered/co-associated loci might simultaneously improve multiple targeted traits.
Linkage analysis and GWAS are two complementary approaches commonly used to dissect the genetic architecture of traits of interest (Lu et al. 2010), and consistency between the QTLs from the linkage mapping and the associated loci from the GWAS could provide cross-validation (Zhang et al. 2017). In the present study, a loci of PZE-101066560 on chromosome 1 was identified as being associated with KW and HKW by GWAS, and this locus was found co-localized within the QTL interval of qKW1-3 (flanked by PZE-101066217 and SYN29479), associated with KW by linkage mapping. In this region, the QTLs for kernel width and kernel weight were identified in previous studies (Li et al. 2013, Liu et al. 2014). Twenty-three protein-encoding genes were located in the genomic region of qKW1-3, of which five were located in the LD block of PZE-101066560. According to the B73 reference genome sequence Version 5b.60 and the gene annotation data available at MaizeGDB, two annotated genes, GRMZM5G824629 (encoding a ubiquitin-conjugating enzyme) and GRMZM2G011483 (encoding a cyclin-related protein), were the most likely candidate genes for kernel width and kernel weight. In rice, GW2 and GW5 are involved in the ubiquitin-proteasome pathway that regulates grain width and weight (Song et al. 2007, Weng et al. 2008). The ubiquitin pathway may also play critical roles in grain development in other plant species (Li et al. 2008, Song et al. 2007). Kernel size might be determined by the stretching ability of the kernel, both longitudinally and latitudinally, during cell division, which affects the final endosperm and embryo sizes (Kesavan et al. 2013, Li et al. 2016). In rice, the gene/QTL GL3.1 that controls grain size and yield might also be involved in cell cycle regulation, suggesting a new mechanism for the regulation of grain size and yield that is driven by cell cycle progression (Qi et al. 2012). The candidate gene GRMZM2G011483 revealed in the present study might be involved in the same pathway for grain size control. These results provide basis for further research into the genetic mechanisms of maize kernel-related traits and for a MAS to increase kernel size in maize yield improvement programs. The fine mapping of these QTLs and further studies on the molecular functions of these potential candidate genes will provide more insights into the underlying genetic and molecular mechanisms of maize grain development, which could be used for maize grain yield improvement.
This work was funded by the National Key Technology Research and Development Program of MOST (2016YFD0100303), the China Postdoctoral Science Foundation (2016M591935), the Jiangsu Planned Projects for Postdoctoral Research Funds (1501116B), the Six Major Talent Project of Jiangsu Province, China (2016-NY-143), the Scientific and Technological Project of Jiangsu Province, China (BE2018325), the Jiangsu Agriculture Science and Technology Innovation Fund (CX(17)2013), and the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD).
